Creation of the High Quality Real 3D Image Representation from the Observe Defective Data.

某些类型的激光测距仪可以同时测量距离和颜色数据，通常用于获取室外景物的三维结构。然而，激光测距仪并不能给出建筑物等目标物体的完美距离信息，各种因素导致测距数据存在严重缺陷。提出了一种基于区域分割的缺陷检测方案，并将非线性时间演化方法应用于距离数据缺陷区域的修复。在缺陷检测方面，我们对观测数据进行距离和颜色分割，将观测数据分为几个区域，分别对应建筑物、天空、地面等。根据分割结果，确定缺陷区域。给定缺陷区域，它们的距离数据将从其邻近区域的观测数据中修复。为此，我们对Bertalmio等人先前开发的用于强度图像缺陷修复的基于传输的补图算法进行了改进，针对距离数据的缺陷修复，构建了一种新的缺陷修复方法，该方法将基于传输的补图和距离数据的每个采样点的原始观测方向上的数据投影组成的交织顺序更新应用于由观测距离数据转换成的三维点数据。利用人为损坏的测试范围数据和真实观测范围数据进行的性能评价表明，我们的修复方法在定量性能和主观修复质量上都优于现有的修复方法。

Some types of laser range scanner can measure range and color data simultaneously, and are often used to acquire 3D structure of outdoor scenery. However, unfortunately a laser range scanner cannot give us perfect range information about target objects such as buildings, and various factors incur critical defects of range data. We present a defect detection scheme based on the region segmentation using observed range-and-color image data, and apply a nonlinear time-evolution method to the repair of defect regions of range data. As to the defect detection, performing the range-andcolor segmentation, we divide observed data into several regions corresponding to buildings, the sky, the ground and so on. Using the segmentation results, we determine defect regions. Given defect regions, their range data will be repaired from observed data in their neighborhoods. For that purpose, reforming the transportation-based inpainting algorithm, previously developed for the defect repair of an intensity image by Bertalmio and others, for the defect repair of range data, we construct a new defect-repair method that applies the interleaved sequential updates, composed of the transportationbased inpainting and the data projection onto the original viewing direction of each sampling point of range data, to 3D point data converted from observed range data. The performance evaluations using artificially damaged test range data and really observed range data demonstrate that our repair method outperforms the existing repair methods both in quantitative performance and in subjective repair quality.